xref: /openbmc/linux/fs/inode.c (revision a09d2831)
1 /*
2  * linux/fs/inode.c
3  *
4  * (C) 1997 Linus Torvalds
5  */
6 
7 #include <linux/fs.h>
8 #include <linux/mm.h>
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/quotaops.h>
12 #include <linux/slab.h>
13 #include <linux/writeback.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/rwsem.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 #include <linux/fsnotify.h>
26 #include <linux/mount.h>
27 #include <linux/async.h>
28 #include <linux/posix_acl.h>
29 
30 /*
31  * This is needed for the following functions:
32  *  - inode_has_buffers
33  *  - invalidate_inode_buffers
34  *  - invalidate_bdev
35  *
36  * FIXME: remove all knowledge of the buffer layer from this file
37  */
38 #include <linux/buffer_head.h>
39 
40 /*
41  * New inode.c implementation.
42  *
43  * This implementation has the basic premise of trying
44  * to be extremely low-overhead and SMP-safe, yet be
45  * simple enough to be "obviously correct".
46  *
47  * Famous last words.
48  */
49 
50 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 
52 /* #define INODE_PARANOIA 1 */
53 /* #define INODE_DEBUG 1 */
54 
55 /*
56  * Inode lookup is no longer as critical as it used to be:
57  * most of the lookups are going to be through the dcache.
58  */
59 #define I_HASHBITS	i_hash_shift
60 #define I_HASHMASK	i_hash_mask
61 
62 static unsigned int i_hash_mask __read_mostly;
63 static unsigned int i_hash_shift __read_mostly;
64 
65 /*
66  * Each inode can be on two separate lists. One is
67  * the hash list of the inode, used for lookups. The
68  * other linked list is the "type" list:
69  *  "in_use" - valid inode, i_count > 0, i_nlink > 0
70  *  "dirty"  - as "in_use" but also dirty
71  *  "unused" - valid inode, i_count = 0
72  *
73  * A "dirty" list is maintained for each super block,
74  * allowing for low-overhead inode sync() operations.
75  */
76 
77 LIST_HEAD(inode_in_use);
78 LIST_HEAD(inode_unused);
79 static struct hlist_head *inode_hashtable __read_mostly;
80 
81 /*
82  * A simple spinlock to protect the list manipulations.
83  *
84  * NOTE! You also have to own the lock if you change
85  * the i_state of an inode while it is in use..
86  */
87 DEFINE_SPINLOCK(inode_lock);
88 
89 /*
90  * iprune_sem provides exclusion between the kswapd or try_to_free_pages
91  * icache shrinking path, and the umount path.  Without this exclusion,
92  * by the time prune_icache calls iput for the inode whose pages it has
93  * been invalidating, or by the time it calls clear_inode & destroy_inode
94  * from its final dispose_list, the struct super_block they refer to
95  * (for inode->i_sb->s_op) may already have been freed and reused.
96  *
97  * We make this an rwsem because the fastpath is icache shrinking. In
98  * some cases a filesystem may be doing a significant amount of work in
99  * its inode reclaim code, so this should improve parallelism.
100  */
101 static DECLARE_RWSEM(iprune_sem);
102 
103 /*
104  * Statistics gathering..
105  */
106 struct inodes_stat_t inodes_stat;
107 
108 static struct kmem_cache *inode_cachep __read_mostly;
109 
110 static void wake_up_inode(struct inode *inode)
111 {
112 	/*
113 	 * Prevent speculative execution through spin_unlock(&inode_lock);
114 	 */
115 	smp_mb();
116 	wake_up_bit(&inode->i_state, __I_NEW);
117 }
118 
119 /**
120  * inode_init_always - perform inode structure intialisation
121  * @sb: superblock inode belongs to
122  * @inode: inode to initialise
123  *
124  * These are initializations that need to be done on every inode
125  * allocation as the fields are not initialised by slab allocation.
126  */
127 int inode_init_always(struct super_block *sb, struct inode *inode)
128 {
129 	static const struct address_space_operations empty_aops;
130 	static const struct inode_operations empty_iops;
131 	static const struct file_operations empty_fops;
132 	struct address_space *const mapping = &inode->i_data;
133 
134 	inode->i_sb = sb;
135 	inode->i_blkbits = sb->s_blocksize_bits;
136 	inode->i_flags = 0;
137 	atomic_set(&inode->i_count, 1);
138 	inode->i_op = &empty_iops;
139 	inode->i_fop = &empty_fops;
140 	inode->i_nlink = 1;
141 	inode->i_uid = 0;
142 	inode->i_gid = 0;
143 	atomic_set(&inode->i_writecount, 0);
144 	inode->i_size = 0;
145 	inode->i_blocks = 0;
146 	inode->i_bytes = 0;
147 	inode->i_generation = 0;
148 #ifdef CONFIG_QUOTA
149 	memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
150 #endif
151 	inode->i_pipe = NULL;
152 	inode->i_bdev = NULL;
153 	inode->i_cdev = NULL;
154 	inode->i_rdev = 0;
155 	inode->dirtied_when = 0;
156 
157 	if (security_inode_alloc(inode))
158 		goto out;
159 	spin_lock_init(&inode->i_lock);
160 	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
161 
162 	mutex_init(&inode->i_mutex);
163 	lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
164 
165 	init_rwsem(&inode->i_alloc_sem);
166 	lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);
167 
168 	mapping->a_ops = &empty_aops;
169 	mapping->host = inode;
170 	mapping->flags = 0;
171 	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
172 	mapping->assoc_mapping = NULL;
173 	mapping->backing_dev_info = &default_backing_dev_info;
174 	mapping->writeback_index = 0;
175 
176 	/*
177 	 * If the block_device provides a backing_dev_info for client
178 	 * inodes then use that.  Otherwise the inode share the bdev's
179 	 * backing_dev_info.
180 	 */
181 	if (sb->s_bdev) {
182 		struct backing_dev_info *bdi;
183 
184 		bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
185 		mapping->backing_dev_info = bdi;
186 	}
187 	inode->i_private = NULL;
188 	inode->i_mapping = mapping;
189 #ifdef CONFIG_FS_POSIX_ACL
190 	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
191 #endif
192 
193 #ifdef CONFIG_FSNOTIFY
194 	inode->i_fsnotify_mask = 0;
195 #endif
196 
197 	return 0;
198 out:
199 	return -ENOMEM;
200 }
201 EXPORT_SYMBOL(inode_init_always);
202 
203 static struct inode *alloc_inode(struct super_block *sb)
204 {
205 	struct inode *inode;
206 
207 	if (sb->s_op->alloc_inode)
208 		inode = sb->s_op->alloc_inode(sb);
209 	else
210 		inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
211 
212 	if (!inode)
213 		return NULL;
214 
215 	if (unlikely(inode_init_always(sb, inode))) {
216 		if (inode->i_sb->s_op->destroy_inode)
217 			inode->i_sb->s_op->destroy_inode(inode);
218 		else
219 			kmem_cache_free(inode_cachep, inode);
220 		return NULL;
221 	}
222 
223 	return inode;
224 }
225 
226 void __destroy_inode(struct inode *inode)
227 {
228 	BUG_ON(inode_has_buffers(inode));
229 	security_inode_free(inode);
230 	fsnotify_inode_delete(inode);
231 #ifdef CONFIG_FS_POSIX_ACL
232 	if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
233 		posix_acl_release(inode->i_acl);
234 	if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
235 		posix_acl_release(inode->i_default_acl);
236 #endif
237 }
238 EXPORT_SYMBOL(__destroy_inode);
239 
240 void destroy_inode(struct inode *inode)
241 {
242 	__destroy_inode(inode);
243 	if (inode->i_sb->s_op->destroy_inode)
244 		inode->i_sb->s_op->destroy_inode(inode);
245 	else
246 		kmem_cache_free(inode_cachep, (inode));
247 }
248 
249 /*
250  * These are initializations that only need to be done
251  * once, because the fields are idempotent across use
252  * of the inode, so let the slab aware of that.
253  */
254 void inode_init_once(struct inode *inode)
255 {
256 	memset(inode, 0, sizeof(*inode));
257 	INIT_HLIST_NODE(&inode->i_hash);
258 	INIT_LIST_HEAD(&inode->i_dentry);
259 	INIT_LIST_HEAD(&inode->i_devices);
260 	INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
261 	spin_lock_init(&inode->i_data.tree_lock);
262 	spin_lock_init(&inode->i_data.i_mmap_lock);
263 	INIT_LIST_HEAD(&inode->i_data.private_list);
264 	spin_lock_init(&inode->i_data.private_lock);
265 	INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
266 	INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
267 	i_size_ordered_init(inode);
268 #ifdef CONFIG_INOTIFY
269 	INIT_LIST_HEAD(&inode->inotify_watches);
270 	mutex_init(&inode->inotify_mutex);
271 #endif
272 #ifdef CONFIG_FSNOTIFY
273 	INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
274 #endif
275 }
276 EXPORT_SYMBOL(inode_init_once);
277 
278 static void init_once(void *foo)
279 {
280 	struct inode *inode = (struct inode *) foo;
281 
282 	inode_init_once(inode);
283 }
284 
285 /*
286  * inode_lock must be held
287  */
288 void __iget(struct inode *inode)
289 {
290 	if (atomic_read(&inode->i_count)) {
291 		atomic_inc(&inode->i_count);
292 		return;
293 	}
294 	atomic_inc(&inode->i_count);
295 	if (!(inode->i_state & (I_DIRTY|I_SYNC)))
296 		list_move(&inode->i_list, &inode_in_use);
297 	inodes_stat.nr_unused--;
298 }
299 
300 /**
301  * clear_inode - clear an inode
302  * @inode: inode to clear
303  *
304  * This is called by the filesystem to tell us
305  * that the inode is no longer useful. We just
306  * terminate it with extreme prejudice.
307  */
308 void clear_inode(struct inode *inode)
309 {
310 	might_sleep();
311 	invalidate_inode_buffers(inode);
312 
313 	BUG_ON(inode->i_data.nrpages);
314 	BUG_ON(!(inode->i_state & I_FREEING));
315 	BUG_ON(inode->i_state & I_CLEAR);
316 	inode_sync_wait(inode);
317 	vfs_dq_drop(inode);
318 	if (inode->i_sb->s_op->clear_inode)
319 		inode->i_sb->s_op->clear_inode(inode);
320 	if (S_ISBLK(inode->i_mode) && inode->i_bdev)
321 		bd_forget(inode);
322 	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
323 		cd_forget(inode);
324 	inode->i_state = I_CLEAR;
325 }
326 EXPORT_SYMBOL(clear_inode);
327 
328 /*
329  * dispose_list - dispose of the contents of a local list
330  * @head: the head of the list to free
331  *
332  * Dispose-list gets a local list with local inodes in it, so it doesn't
333  * need to worry about list corruption and SMP locks.
334  */
335 static void dispose_list(struct list_head *head)
336 {
337 	int nr_disposed = 0;
338 
339 	while (!list_empty(head)) {
340 		struct inode *inode;
341 
342 		inode = list_first_entry(head, struct inode, i_list);
343 		list_del(&inode->i_list);
344 
345 		if (inode->i_data.nrpages)
346 			truncate_inode_pages(&inode->i_data, 0);
347 		clear_inode(inode);
348 
349 		spin_lock(&inode_lock);
350 		hlist_del_init(&inode->i_hash);
351 		list_del_init(&inode->i_sb_list);
352 		spin_unlock(&inode_lock);
353 
354 		wake_up_inode(inode);
355 		destroy_inode(inode);
356 		nr_disposed++;
357 	}
358 	spin_lock(&inode_lock);
359 	inodes_stat.nr_inodes -= nr_disposed;
360 	spin_unlock(&inode_lock);
361 }
362 
363 /*
364  * Invalidate all inodes for a device.
365  */
366 static int invalidate_list(struct list_head *head, struct list_head *dispose)
367 {
368 	struct list_head *next;
369 	int busy = 0, count = 0;
370 
371 	next = head->next;
372 	for (;;) {
373 		struct list_head *tmp = next;
374 		struct inode *inode;
375 
376 		/*
377 		 * We can reschedule here without worrying about the list's
378 		 * consistency because the per-sb list of inodes must not
379 		 * change during umount anymore, and because iprune_sem keeps
380 		 * shrink_icache_memory() away.
381 		 */
382 		cond_resched_lock(&inode_lock);
383 
384 		next = next->next;
385 		if (tmp == head)
386 			break;
387 		inode = list_entry(tmp, struct inode, i_sb_list);
388 		if (inode->i_state & I_NEW)
389 			continue;
390 		invalidate_inode_buffers(inode);
391 		if (!atomic_read(&inode->i_count)) {
392 			list_move(&inode->i_list, dispose);
393 			WARN_ON(inode->i_state & I_NEW);
394 			inode->i_state |= I_FREEING;
395 			count++;
396 			continue;
397 		}
398 		busy = 1;
399 	}
400 	/* only unused inodes may be cached with i_count zero */
401 	inodes_stat.nr_unused -= count;
402 	return busy;
403 }
404 
405 /**
406  *	invalidate_inodes	- discard the inodes on a device
407  *	@sb: superblock
408  *
409  *	Discard all of the inodes for a given superblock. If the discard
410  *	fails because there are busy inodes then a non zero value is returned.
411  *	If the discard is successful all the inodes have been discarded.
412  */
413 int invalidate_inodes(struct super_block *sb)
414 {
415 	int busy;
416 	LIST_HEAD(throw_away);
417 
418 	down_write(&iprune_sem);
419 	spin_lock(&inode_lock);
420 	inotify_unmount_inodes(&sb->s_inodes);
421 	fsnotify_unmount_inodes(&sb->s_inodes);
422 	busy = invalidate_list(&sb->s_inodes, &throw_away);
423 	spin_unlock(&inode_lock);
424 
425 	dispose_list(&throw_away);
426 	up_write(&iprune_sem);
427 
428 	return busy;
429 }
430 EXPORT_SYMBOL(invalidate_inodes);
431 
432 static int can_unuse(struct inode *inode)
433 {
434 	if (inode->i_state)
435 		return 0;
436 	if (inode_has_buffers(inode))
437 		return 0;
438 	if (atomic_read(&inode->i_count))
439 		return 0;
440 	if (inode->i_data.nrpages)
441 		return 0;
442 	return 1;
443 }
444 
445 /*
446  * Scan `goal' inodes on the unused list for freeable ones. They are moved to
447  * a temporary list and then are freed outside inode_lock by dispose_list().
448  *
449  * Any inodes which are pinned purely because of attached pagecache have their
450  * pagecache removed.  We expect the final iput() on that inode to add it to
451  * the front of the inode_unused list.  So look for it there and if the
452  * inode is still freeable, proceed.  The right inode is found 99.9% of the
453  * time in testing on a 4-way.
454  *
455  * If the inode has metadata buffers attached to mapping->private_list then
456  * try to remove them.
457  */
458 static void prune_icache(int nr_to_scan)
459 {
460 	LIST_HEAD(freeable);
461 	int nr_pruned = 0;
462 	int nr_scanned;
463 	unsigned long reap = 0;
464 
465 	down_read(&iprune_sem);
466 	spin_lock(&inode_lock);
467 	for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
468 		struct inode *inode;
469 
470 		if (list_empty(&inode_unused))
471 			break;
472 
473 		inode = list_entry(inode_unused.prev, struct inode, i_list);
474 
475 		if (inode->i_state || atomic_read(&inode->i_count)) {
476 			list_move(&inode->i_list, &inode_unused);
477 			continue;
478 		}
479 		if (inode_has_buffers(inode) || inode->i_data.nrpages) {
480 			__iget(inode);
481 			spin_unlock(&inode_lock);
482 			if (remove_inode_buffers(inode))
483 				reap += invalidate_mapping_pages(&inode->i_data,
484 								0, -1);
485 			iput(inode);
486 			spin_lock(&inode_lock);
487 
488 			if (inode != list_entry(inode_unused.next,
489 						struct inode, i_list))
490 				continue;	/* wrong inode or list_empty */
491 			if (!can_unuse(inode))
492 				continue;
493 		}
494 		list_move(&inode->i_list, &freeable);
495 		WARN_ON(inode->i_state & I_NEW);
496 		inode->i_state |= I_FREEING;
497 		nr_pruned++;
498 	}
499 	inodes_stat.nr_unused -= nr_pruned;
500 	if (current_is_kswapd())
501 		__count_vm_events(KSWAPD_INODESTEAL, reap);
502 	else
503 		__count_vm_events(PGINODESTEAL, reap);
504 	spin_unlock(&inode_lock);
505 
506 	dispose_list(&freeable);
507 	up_read(&iprune_sem);
508 }
509 
510 /*
511  * shrink_icache_memory() will attempt to reclaim some unused inodes.  Here,
512  * "unused" means that no dentries are referring to the inodes: the files are
513  * not open and the dcache references to those inodes have already been
514  * reclaimed.
515  *
516  * This function is passed the number of inodes to scan, and it returns the
517  * total number of remaining possibly-reclaimable inodes.
518  */
519 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
520 {
521 	if (nr) {
522 		/*
523 		 * Nasty deadlock avoidance.  We may hold various FS locks,
524 		 * and we don't want to recurse into the FS that called us
525 		 * in clear_inode() and friends..
526 		 */
527 		if (!(gfp_mask & __GFP_FS))
528 			return -1;
529 		prune_icache(nr);
530 	}
531 	return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
532 }
533 
534 static struct shrinker icache_shrinker = {
535 	.shrink = shrink_icache_memory,
536 	.seeks = DEFAULT_SEEKS,
537 };
538 
539 static void __wait_on_freeing_inode(struct inode *inode);
540 /*
541  * Called with the inode lock held.
542  * NOTE: we are not increasing the inode-refcount, you must call __iget()
543  * by hand after calling find_inode now! This simplifies iunique and won't
544  * add any additional branch in the common code.
545  */
546 static struct inode *find_inode(struct super_block *sb,
547 				struct hlist_head *head,
548 				int (*test)(struct inode *, void *),
549 				void *data)
550 {
551 	struct hlist_node *node;
552 	struct inode *inode = NULL;
553 
554 repeat:
555 	hlist_for_each_entry(inode, node, head, i_hash) {
556 		if (inode->i_sb != sb)
557 			continue;
558 		if (!test(inode, data))
559 			continue;
560 		if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
561 			__wait_on_freeing_inode(inode);
562 			goto repeat;
563 		}
564 		break;
565 	}
566 	return node ? inode : NULL;
567 }
568 
569 /*
570  * find_inode_fast is the fast path version of find_inode, see the comment at
571  * iget_locked for details.
572  */
573 static struct inode *find_inode_fast(struct super_block *sb,
574 				struct hlist_head *head, unsigned long ino)
575 {
576 	struct hlist_node *node;
577 	struct inode *inode = NULL;
578 
579 repeat:
580 	hlist_for_each_entry(inode, node, head, i_hash) {
581 		if (inode->i_ino != ino)
582 			continue;
583 		if (inode->i_sb != sb)
584 			continue;
585 		if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
586 			__wait_on_freeing_inode(inode);
587 			goto repeat;
588 		}
589 		break;
590 	}
591 	return node ? inode : NULL;
592 }
593 
594 static unsigned long hash(struct super_block *sb, unsigned long hashval)
595 {
596 	unsigned long tmp;
597 
598 	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
599 			L1_CACHE_BYTES;
600 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
601 	return tmp & I_HASHMASK;
602 }
603 
604 static inline void
605 __inode_add_to_lists(struct super_block *sb, struct hlist_head *head,
606 			struct inode *inode)
607 {
608 	inodes_stat.nr_inodes++;
609 	list_add(&inode->i_list, &inode_in_use);
610 	list_add(&inode->i_sb_list, &sb->s_inodes);
611 	if (head)
612 		hlist_add_head(&inode->i_hash, head);
613 }
614 
615 /**
616  * inode_add_to_lists - add a new inode to relevant lists
617  * @sb: superblock inode belongs to
618  * @inode: inode to mark in use
619  *
620  * When an inode is allocated it needs to be accounted for, added to the in use
621  * list, the owning superblock and the inode hash. This needs to be done under
622  * the inode_lock, so export a function to do this rather than the inode lock
623  * itself. We calculate the hash list to add to here so it is all internal
624  * which requires the caller to have already set up the inode number in the
625  * inode to add.
626  */
627 void inode_add_to_lists(struct super_block *sb, struct inode *inode)
628 {
629 	struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino);
630 
631 	spin_lock(&inode_lock);
632 	__inode_add_to_lists(sb, head, inode);
633 	spin_unlock(&inode_lock);
634 }
635 EXPORT_SYMBOL_GPL(inode_add_to_lists);
636 
637 /**
638  *	new_inode 	- obtain an inode
639  *	@sb: superblock
640  *
641  *	Allocates a new inode for given superblock. The default gfp_mask
642  *	for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
643  *	If HIGHMEM pages are unsuitable or it is known that pages allocated
644  *	for the page cache are not reclaimable or migratable,
645  *	mapping_set_gfp_mask() must be called with suitable flags on the
646  *	newly created inode's mapping
647  *
648  */
649 struct inode *new_inode(struct super_block *sb)
650 {
651 	/*
652 	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
653 	 * error if st_ino won't fit in target struct field. Use 32bit counter
654 	 * here to attempt to avoid that.
655 	 */
656 	static unsigned int last_ino;
657 	struct inode *inode;
658 
659 	spin_lock_prefetch(&inode_lock);
660 
661 	inode = alloc_inode(sb);
662 	if (inode) {
663 		spin_lock(&inode_lock);
664 		__inode_add_to_lists(sb, NULL, inode);
665 		inode->i_ino = ++last_ino;
666 		inode->i_state = 0;
667 		spin_unlock(&inode_lock);
668 	}
669 	return inode;
670 }
671 EXPORT_SYMBOL(new_inode);
672 
673 void unlock_new_inode(struct inode *inode)
674 {
675 #ifdef CONFIG_DEBUG_LOCK_ALLOC
676 	if (inode->i_mode & S_IFDIR) {
677 		struct file_system_type *type = inode->i_sb->s_type;
678 
679 		/* Set new key only if filesystem hasn't already changed it */
680 		if (!lockdep_match_class(&inode->i_mutex,
681 		    &type->i_mutex_key)) {
682 			/*
683 			 * ensure nobody is actually holding i_mutex
684 			 */
685 			mutex_destroy(&inode->i_mutex);
686 			mutex_init(&inode->i_mutex);
687 			lockdep_set_class(&inode->i_mutex,
688 					  &type->i_mutex_dir_key);
689 		}
690 	}
691 #endif
692 	/*
693 	 * This is special!  We do not need the spinlock when clearing I_NEW,
694 	 * because we're guaranteed that nobody else tries to do anything about
695 	 * the state of the inode when it is locked, as we just created it (so
696 	 * there can be no old holders that haven't tested I_NEW).
697 	 * However we must emit the memory barrier so that other CPUs reliably
698 	 * see the clearing of I_NEW after the other inode initialisation has
699 	 * completed.
700 	 */
701 	smp_mb();
702 	WARN_ON(!(inode->i_state & I_NEW));
703 	inode->i_state &= ~I_NEW;
704 	wake_up_inode(inode);
705 }
706 EXPORT_SYMBOL(unlock_new_inode);
707 
708 /*
709  * This is called without the inode lock held.. Be careful.
710  *
711  * We no longer cache the sb_flags in i_flags - see fs.h
712  *	-- rmk@arm.uk.linux.org
713  */
714 static struct inode *get_new_inode(struct super_block *sb,
715 				struct hlist_head *head,
716 				int (*test)(struct inode *, void *),
717 				int (*set)(struct inode *, void *),
718 				void *data)
719 {
720 	struct inode *inode;
721 
722 	inode = alloc_inode(sb);
723 	if (inode) {
724 		struct inode *old;
725 
726 		spin_lock(&inode_lock);
727 		/* We released the lock, so.. */
728 		old = find_inode(sb, head, test, data);
729 		if (!old) {
730 			if (set(inode, data))
731 				goto set_failed;
732 
733 			__inode_add_to_lists(sb, head, inode);
734 			inode->i_state = I_NEW;
735 			spin_unlock(&inode_lock);
736 
737 			/* Return the locked inode with I_NEW set, the
738 			 * caller is responsible for filling in the contents
739 			 */
740 			return inode;
741 		}
742 
743 		/*
744 		 * Uhhuh, somebody else created the same inode under
745 		 * us. Use the old inode instead of the one we just
746 		 * allocated.
747 		 */
748 		__iget(old);
749 		spin_unlock(&inode_lock);
750 		destroy_inode(inode);
751 		inode = old;
752 		wait_on_inode(inode);
753 	}
754 	return inode;
755 
756 set_failed:
757 	spin_unlock(&inode_lock);
758 	destroy_inode(inode);
759 	return NULL;
760 }
761 
762 /*
763  * get_new_inode_fast is the fast path version of get_new_inode, see the
764  * comment at iget_locked for details.
765  */
766 static struct inode *get_new_inode_fast(struct super_block *sb,
767 				struct hlist_head *head, unsigned long ino)
768 {
769 	struct inode *inode;
770 
771 	inode = alloc_inode(sb);
772 	if (inode) {
773 		struct inode *old;
774 
775 		spin_lock(&inode_lock);
776 		/* We released the lock, so.. */
777 		old = find_inode_fast(sb, head, ino);
778 		if (!old) {
779 			inode->i_ino = ino;
780 			__inode_add_to_lists(sb, head, inode);
781 			inode->i_state = I_NEW;
782 			spin_unlock(&inode_lock);
783 
784 			/* Return the locked inode with I_NEW set, the
785 			 * caller is responsible for filling in the contents
786 			 */
787 			return inode;
788 		}
789 
790 		/*
791 		 * Uhhuh, somebody else created the same inode under
792 		 * us. Use the old inode instead of the one we just
793 		 * allocated.
794 		 */
795 		__iget(old);
796 		spin_unlock(&inode_lock);
797 		destroy_inode(inode);
798 		inode = old;
799 		wait_on_inode(inode);
800 	}
801 	return inode;
802 }
803 
804 /**
805  *	iunique - get a unique inode number
806  *	@sb: superblock
807  *	@max_reserved: highest reserved inode number
808  *
809  *	Obtain an inode number that is unique on the system for a given
810  *	superblock. This is used by file systems that have no natural
811  *	permanent inode numbering system. An inode number is returned that
812  *	is higher than the reserved limit but unique.
813  *
814  *	BUGS:
815  *	With a large number of inodes live on the file system this function
816  *	currently becomes quite slow.
817  */
818 ino_t iunique(struct super_block *sb, ino_t max_reserved)
819 {
820 	/*
821 	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
822 	 * error if st_ino won't fit in target struct field. Use 32bit counter
823 	 * here to attempt to avoid that.
824 	 */
825 	static unsigned int counter;
826 	struct inode *inode;
827 	struct hlist_head *head;
828 	ino_t res;
829 
830 	spin_lock(&inode_lock);
831 	do {
832 		if (counter <= max_reserved)
833 			counter = max_reserved + 1;
834 		res = counter++;
835 		head = inode_hashtable + hash(sb, res);
836 		inode = find_inode_fast(sb, head, res);
837 	} while (inode != NULL);
838 	spin_unlock(&inode_lock);
839 
840 	return res;
841 }
842 EXPORT_SYMBOL(iunique);
843 
844 struct inode *igrab(struct inode *inode)
845 {
846 	spin_lock(&inode_lock);
847 	if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)))
848 		__iget(inode);
849 	else
850 		/*
851 		 * Handle the case where s_op->clear_inode is not been
852 		 * called yet, and somebody is calling igrab
853 		 * while the inode is getting freed.
854 		 */
855 		inode = NULL;
856 	spin_unlock(&inode_lock);
857 	return inode;
858 }
859 EXPORT_SYMBOL(igrab);
860 
861 /**
862  * ifind - internal function, you want ilookup5() or iget5().
863  * @sb:		super block of file system to search
864  * @head:       the head of the list to search
865  * @test:	callback used for comparisons between inodes
866  * @data:	opaque data pointer to pass to @test
867  * @wait:	if true wait for the inode to be unlocked, if false do not
868  *
869  * ifind() searches for the inode specified by @data in the inode
870  * cache. This is a generalized version of ifind_fast() for file systems where
871  * the inode number is not sufficient for unique identification of an inode.
872  *
873  * If the inode is in the cache, the inode is returned with an incremented
874  * reference count.
875  *
876  * Otherwise NULL is returned.
877  *
878  * Note, @test is called with the inode_lock held, so can't sleep.
879  */
880 static struct inode *ifind(struct super_block *sb,
881 		struct hlist_head *head, int (*test)(struct inode *, void *),
882 		void *data, const int wait)
883 {
884 	struct inode *inode;
885 
886 	spin_lock(&inode_lock);
887 	inode = find_inode(sb, head, test, data);
888 	if (inode) {
889 		__iget(inode);
890 		spin_unlock(&inode_lock);
891 		if (likely(wait))
892 			wait_on_inode(inode);
893 		return inode;
894 	}
895 	spin_unlock(&inode_lock);
896 	return NULL;
897 }
898 
899 /**
900  * ifind_fast - internal function, you want ilookup() or iget().
901  * @sb:		super block of file system to search
902  * @head:       head of the list to search
903  * @ino:	inode number to search for
904  *
905  * ifind_fast() searches for the inode @ino in the inode cache. This is for
906  * file systems where the inode number is sufficient for unique identification
907  * of an inode.
908  *
909  * If the inode is in the cache, the inode is returned with an incremented
910  * reference count.
911  *
912  * Otherwise NULL is returned.
913  */
914 static struct inode *ifind_fast(struct super_block *sb,
915 		struct hlist_head *head, unsigned long ino)
916 {
917 	struct inode *inode;
918 
919 	spin_lock(&inode_lock);
920 	inode = find_inode_fast(sb, head, ino);
921 	if (inode) {
922 		__iget(inode);
923 		spin_unlock(&inode_lock);
924 		wait_on_inode(inode);
925 		return inode;
926 	}
927 	spin_unlock(&inode_lock);
928 	return NULL;
929 }
930 
931 /**
932  * ilookup5_nowait - search for an inode in the inode cache
933  * @sb:		super block of file system to search
934  * @hashval:	hash value (usually inode number) to search for
935  * @test:	callback used for comparisons between inodes
936  * @data:	opaque data pointer to pass to @test
937  *
938  * ilookup5() uses ifind() to search for the inode specified by @hashval and
939  * @data in the inode cache. This is a generalized version of ilookup() for
940  * file systems where the inode number is not sufficient for unique
941  * identification of an inode.
942  *
943  * If the inode is in the cache, the inode is returned with an incremented
944  * reference count.  Note, the inode lock is not waited upon so you have to be
945  * very careful what you do with the returned inode.  You probably should be
946  * using ilookup5() instead.
947  *
948  * Otherwise NULL is returned.
949  *
950  * Note, @test is called with the inode_lock held, so can't sleep.
951  */
952 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
953 		int (*test)(struct inode *, void *), void *data)
954 {
955 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
956 
957 	return ifind(sb, head, test, data, 0);
958 }
959 EXPORT_SYMBOL(ilookup5_nowait);
960 
961 /**
962  * ilookup5 - search for an inode in the inode cache
963  * @sb:		super block of file system to search
964  * @hashval:	hash value (usually inode number) to search for
965  * @test:	callback used for comparisons between inodes
966  * @data:	opaque data pointer to pass to @test
967  *
968  * ilookup5() uses ifind() to search for the inode specified by @hashval and
969  * @data in the inode cache. This is a generalized version of ilookup() for
970  * file systems where the inode number is not sufficient for unique
971  * identification of an inode.
972  *
973  * If the inode is in the cache, the inode lock is waited upon and the inode is
974  * returned with an incremented reference count.
975  *
976  * Otherwise NULL is returned.
977  *
978  * Note, @test is called with the inode_lock held, so can't sleep.
979  */
980 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
981 		int (*test)(struct inode *, void *), void *data)
982 {
983 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
984 
985 	return ifind(sb, head, test, data, 1);
986 }
987 EXPORT_SYMBOL(ilookup5);
988 
989 /**
990  * ilookup - search for an inode in the inode cache
991  * @sb:		super block of file system to search
992  * @ino:	inode number to search for
993  *
994  * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
995  * This is for file systems where the inode number is sufficient for unique
996  * identification of an inode.
997  *
998  * If the inode is in the cache, the inode is returned with an incremented
999  * reference count.
1000  *
1001  * Otherwise NULL is returned.
1002  */
1003 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1004 {
1005 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1006 
1007 	return ifind_fast(sb, head, ino);
1008 }
1009 EXPORT_SYMBOL(ilookup);
1010 
1011 /**
1012  * iget5_locked - obtain an inode from a mounted file system
1013  * @sb:		super block of file system
1014  * @hashval:	hash value (usually inode number) to get
1015  * @test:	callback used for comparisons between inodes
1016  * @set:	callback used to initialize a new struct inode
1017  * @data:	opaque data pointer to pass to @test and @set
1018  *
1019  * iget5_locked() uses ifind() to search for the inode specified by @hashval
1020  * and @data in the inode cache and if present it is returned with an increased
1021  * reference count. This is a generalized version of iget_locked() for file
1022  * systems where the inode number is not sufficient for unique identification
1023  * of an inode.
1024  *
1025  * If the inode is not in cache, get_new_inode() is called to allocate a new
1026  * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1027  * file system gets to fill it in before unlocking it via unlock_new_inode().
1028  *
1029  * Note both @test and @set are called with the inode_lock held, so can't sleep.
1030  */
1031 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1032 		int (*test)(struct inode *, void *),
1033 		int (*set)(struct inode *, void *), void *data)
1034 {
1035 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1036 	struct inode *inode;
1037 
1038 	inode = ifind(sb, head, test, data, 1);
1039 	if (inode)
1040 		return inode;
1041 	/*
1042 	 * get_new_inode() will do the right thing, re-trying the search
1043 	 * in case it had to block at any point.
1044 	 */
1045 	return get_new_inode(sb, head, test, set, data);
1046 }
1047 EXPORT_SYMBOL(iget5_locked);
1048 
1049 /**
1050  * iget_locked - obtain an inode from a mounted file system
1051  * @sb:		super block of file system
1052  * @ino:	inode number to get
1053  *
1054  * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1055  * the inode cache and if present it is returned with an increased reference
1056  * count. This is for file systems where the inode number is sufficient for
1057  * unique identification of an inode.
1058  *
1059  * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1060  * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1061  * The file system gets to fill it in before unlocking it via
1062  * unlock_new_inode().
1063  */
1064 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1065 {
1066 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1067 	struct inode *inode;
1068 
1069 	inode = ifind_fast(sb, head, ino);
1070 	if (inode)
1071 		return inode;
1072 	/*
1073 	 * get_new_inode_fast() will do the right thing, re-trying the search
1074 	 * in case it had to block at any point.
1075 	 */
1076 	return get_new_inode_fast(sb, head, ino);
1077 }
1078 EXPORT_SYMBOL(iget_locked);
1079 
1080 int insert_inode_locked(struct inode *inode)
1081 {
1082 	struct super_block *sb = inode->i_sb;
1083 	ino_t ino = inode->i_ino;
1084 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1085 
1086 	inode->i_state |= I_NEW;
1087 	while (1) {
1088 		struct hlist_node *node;
1089 		struct inode *old = NULL;
1090 		spin_lock(&inode_lock);
1091 		hlist_for_each_entry(old, node, head, i_hash) {
1092 			if (old->i_ino != ino)
1093 				continue;
1094 			if (old->i_sb != sb)
1095 				continue;
1096 			if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1097 				continue;
1098 			break;
1099 		}
1100 		if (likely(!node)) {
1101 			hlist_add_head(&inode->i_hash, head);
1102 			spin_unlock(&inode_lock);
1103 			return 0;
1104 		}
1105 		__iget(old);
1106 		spin_unlock(&inode_lock);
1107 		wait_on_inode(old);
1108 		if (unlikely(!hlist_unhashed(&old->i_hash))) {
1109 			iput(old);
1110 			return -EBUSY;
1111 		}
1112 		iput(old);
1113 	}
1114 }
1115 EXPORT_SYMBOL(insert_inode_locked);
1116 
1117 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1118 		int (*test)(struct inode *, void *), void *data)
1119 {
1120 	struct super_block *sb = inode->i_sb;
1121 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1122 
1123 	inode->i_state |= I_NEW;
1124 
1125 	while (1) {
1126 		struct hlist_node *node;
1127 		struct inode *old = NULL;
1128 
1129 		spin_lock(&inode_lock);
1130 		hlist_for_each_entry(old, node, head, i_hash) {
1131 			if (old->i_sb != sb)
1132 				continue;
1133 			if (!test(old, data))
1134 				continue;
1135 			if (old->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))
1136 				continue;
1137 			break;
1138 		}
1139 		if (likely(!node)) {
1140 			hlist_add_head(&inode->i_hash, head);
1141 			spin_unlock(&inode_lock);
1142 			return 0;
1143 		}
1144 		__iget(old);
1145 		spin_unlock(&inode_lock);
1146 		wait_on_inode(old);
1147 		if (unlikely(!hlist_unhashed(&old->i_hash))) {
1148 			iput(old);
1149 			return -EBUSY;
1150 		}
1151 		iput(old);
1152 	}
1153 }
1154 EXPORT_SYMBOL(insert_inode_locked4);
1155 
1156 /**
1157  *	__insert_inode_hash - hash an inode
1158  *	@inode: unhashed inode
1159  *	@hashval: unsigned long value used to locate this object in the
1160  *		inode_hashtable.
1161  *
1162  *	Add an inode to the inode hash for this superblock.
1163  */
1164 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
1165 {
1166 	struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1167 	spin_lock(&inode_lock);
1168 	hlist_add_head(&inode->i_hash, head);
1169 	spin_unlock(&inode_lock);
1170 }
1171 EXPORT_SYMBOL(__insert_inode_hash);
1172 
1173 /**
1174  *	remove_inode_hash - remove an inode from the hash
1175  *	@inode: inode to unhash
1176  *
1177  *	Remove an inode from the superblock.
1178  */
1179 void remove_inode_hash(struct inode *inode)
1180 {
1181 	spin_lock(&inode_lock);
1182 	hlist_del_init(&inode->i_hash);
1183 	spin_unlock(&inode_lock);
1184 }
1185 EXPORT_SYMBOL(remove_inode_hash);
1186 
1187 /*
1188  * Tell the filesystem that this inode is no longer of any interest and should
1189  * be completely destroyed.
1190  *
1191  * We leave the inode in the inode hash table until *after* the filesystem's
1192  * ->delete_inode completes.  This ensures that an iget (such as nfsd might
1193  * instigate) will always find up-to-date information either in the hash or on
1194  * disk.
1195  *
1196  * I_FREEING is set so that no-one will take a new reference to the inode while
1197  * it is being deleted.
1198  */
1199 void generic_delete_inode(struct inode *inode)
1200 {
1201 	const struct super_operations *op = inode->i_sb->s_op;
1202 
1203 	list_del_init(&inode->i_list);
1204 	list_del_init(&inode->i_sb_list);
1205 	WARN_ON(inode->i_state & I_NEW);
1206 	inode->i_state |= I_FREEING;
1207 	inodes_stat.nr_inodes--;
1208 	spin_unlock(&inode_lock);
1209 
1210 	security_inode_delete(inode);
1211 
1212 	if (op->delete_inode) {
1213 		void (*delete)(struct inode *) = op->delete_inode;
1214 		if (!is_bad_inode(inode))
1215 			vfs_dq_init(inode);
1216 		/* Filesystems implementing their own
1217 		 * s_op->delete_inode are required to call
1218 		 * truncate_inode_pages and clear_inode()
1219 		 * internally */
1220 		delete(inode);
1221 	} else {
1222 		truncate_inode_pages(&inode->i_data, 0);
1223 		clear_inode(inode);
1224 	}
1225 	spin_lock(&inode_lock);
1226 	hlist_del_init(&inode->i_hash);
1227 	spin_unlock(&inode_lock);
1228 	wake_up_inode(inode);
1229 	BUG_ON(inode->i_state != I_CLEAR);
1230 	destroy_inode(inode);
1231 }
1232 EXPORT_SYMBOL(generic_delete_inode);
1233 
1234 /**
1235  *	generic_detach_inode - remove inode from inode lists
1236  *	@inode: inode to remove
1237  *
1238  *	Remove inode from inode lists, write it if it's dirty. This is just an
1239  *	internal VFS helper exported for hugetlbfs. Do not use!
1240  *
1241  *	Returns 1 if inode should be completely destroyed.
1242  */
1243 int generic_detach_inode(struct inode *inode)
1244 {
1245 	struct super_block *sb = inode->i_sb;
1246 
1247 	if (!hlist_unhashed(&inode->i_hash)) {
1248 		if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1249 			list_move(&inode->i_list, &inode_unused);
1250 		inodes_stat.nr_unused++;
1251 		if (sb->s_flags & MS_ACTIVE) {
1252 			spin_unlock(&inode_lock);
1253 			return 0;
1254 		}
1255 		WARN_ON(inode->i_state & I_NEW);
1256 		inode->i_state |= I_WILL_FREE;
1257 		spin_unlock(&inode_lock);
1258 		write_inode_now(inode, 1);
1259 		spin_lock(&inode_lock);
1260 		WARN_ON(inode->i_state & I_NEW);
1261 		inode->i_state &= ~I_WILL_FREE;
1262 		inodes_stat.nr_unused--;
1263 		hlist_del_init(&inode->i_hash);
1264 	}
1265 	list_del_init(&inode->i_list);
1266 	list_del_init(&inode->i_sb_list);
1267 	WARN_ON(inode->i_state & I_NEW);
1268 	inode->i_state |= I_FREEING;
1269 	inodes_stat.nr_inodes--;
1270 	spin_unlock(&inode_lock);
1271 	return 1;
1272 }
1273 EXPORT_SYMBOL_GPL(generic_detach_inode);
1274 
1275 static void generic_forget_inode(struct inode *inode)
1276 {
1277 	if (!generic_detach_inode(inode))
1278 		return;
1279 	if (inode->i_data.nrpages)
1280 		truncate_inode_pages(&inode->i_data, 0);
1281 	clear_inode(inode);
1282 	wake_up_inode(inode);
1283 	destroy_inode(inode);
1284 }
1285 
1286 /*
1287  * Normal UNIX filesystem behaviour: delete the
1288  * inode when the usage count drops to zero, and
1289  * i_nlink is zero.
1290  */
1291 void generic_drop_inode(struct inode *inode)
1292 {
1293 	if (!inode->i_nlink)
1294 		generic_delete_inode(inode);
1295 	else
1296 		generic_forget_inode(inode);
1297 }
1298 EXPORT_SYMBOL_GPL(generic_drop_inode);
1299 
1300 /*
1301  * Called when we're dropping the last reference
1302  * to an inode.
1303  *
1304  * Call the FS "drop()" function, defaulting to
1305  * the legacy UNIX filesystem behaviour..
1306  *
1307  * NOTE! NOTE! NOTE! We're called with the inode lock
1308  * held, and the drop function is supposed to release
1309  * the lock!
1310  */
1311 static inline void iput_final(struct inode *inode)
1312 {
1313 	const struct super_operations *op = inode->i_sb->s_op;
1314 	void (*drop)(struct inode *) = generic_drop_inode;
1315 
1316 	if (op && op->drop_inode)
1317 		drop = op->drop_inode;
1318 	drop(inode);
1319 }
1320 
1321 /**
1322  *	iput	- put an inode
1323  *	@inode: inode to put
1324  *
1325  *	Puts an inode, dropping its usage count. If the inode use count hits
1326  *	zero, the inode is then freed and may also be destroyed.
1327  *
1328  *	Consequently, iput() can sleep.
1329  */
1330 void iput(struct inode *inode)
1331 {
1332 	if (inode) {
1333 		BUG_ON(inode->i_state == I_CLEAR);
1334 
1335 		if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1336 			iput_final(inode);
1337 	}
1338 }
1339 EXPORT_SYMBOL(iput);
1340 
1341 /**
1342  *	bmap	- find a block number in a file
1343  *	@inode: inode of file
1344  *	@block: block to find
1345  *
1346  *	Returns the block number on the device holding the inode that
1347  *	is the disk block number for the block of the file requested.
1348  *	That is, asked for block 4 of inode 1 the function will return the
1349  *	disk block relative to the disk start that holds that block of the
1350  *	file.
1351  */
1352 sector_t bmap(struct inode *inode, sector_t block)
1353 {
1354 	sector_t res = 0;
1355 	if (inode->i_mapping->a_ops->bmap)
1356 		res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1357 	return res;
1358 }
1359 EXPORT_SYMBOL(bmap);
1360 
1361 /*
1362  * With relative atime, only update atime if the previous atime is
1363  * earlier than either the ctime or mtime or if at least a day has
1364  * passed since the last atime update.
1365  */
1366 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1367 			     struct timespec now)
1368 {
1369 
1370 	if (!(mnt->mnt_flags & MNT_RELATIME))
1371 		return 1;
1372 	/*
1373 	 * Is mtime younger than atime? If yes, update atime:
1374 	 */
1375 	if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1376 		return 1;
1377 	/*
1378 	 * Is ctime younger than atime? If yes, update atime:
1379 	 */
1380 	if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1381 		return 1;
1382 
1383 	/*
1384 	 * Is the previous atime value older than a day? If yes,
1385 	 * update atime:
1386 	 */
1387 	if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1388 		return 1;
1389 	/*
1390 	 * Good, we can skip the atime update:
1391 	 */
1392 	return 0;
1393 }
1394 
1395 /**
1396  *	touch_atime	-	update the access time
1397  *	@mnt: mount the inode is accessed on
1398  *	@dentry: dentry accessed
1399  *
1400  *	Update the accessed time on an inode and mark it for writeback.
1401  *	This function automatically handles read only file systems and media,
1402  *	as well as the "noatime" flag and inode specific "noatime" markers.
1403  */
1404 void touch_atime(struct vfsmount *mnt, struct dentry *dentry)
1405 {
1406 	struct inode *inode = dentry->d_inode;
1407 	struct timespec now;
1408 
1409 	if (inode->i_flags & S_NOATIME)
1410 		return;
1411 	if (IS_NOATIME(inode))
1412 		return;
1413 	if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1414 		return;
1415 
1416 	if (mnt->mnt_flags & MNT_NOATIME)
1417 		return;
1418 	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1419 		return;
1420 
1421 	now = current_fs_time(inode->i_sb);
1422 
1423 	if (!relatime_need_update(mnt, inode, now))
1424 		return;
1425 
1426 	if (timespec_equal(&inode->i_atime, &now))
1427 		return;
1428 
1429 	if (mnt_want_write(mnt))
1430 		return;
1431 
1432 	inode->i_atime = now;
1433 	mark_inode_dirty_sync(inode);
1434 	mnt_drop_write(mnt);
1435 }
1436 EXPORT_SYMBOL(touch_atime);
1437 
1438 /**
1439  *	file_update_time	-	update mtime and ctime time
1440  *	@file: file accessed
1441  *
1442  *	Update the mtime and ctime members of an inode and mark the inode
1443  *	for writeback.  Note that this function is meant exclusively for
1444  *	usage in the file write path of filesystems, and filesystems may
1445  *	choose to explicitly ignore update via this function with the
1446  *	S_NOCMTIME inode flag, e.g. for network filesystem where these
1447  *	timestamps are handled by the server.
1448  */
1449 
1450 void file_update_time(struct file *file)
1451 {
1452 	struct inode *inode = file->f_path.dentry->d_inode;
1453 	struct timespec now;
1454 	enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
1455 
1456 	/* First try to exhaust all avenues to not sync */
1457 	if (IS_NOCMTIME(inode))
1458 		return;
1459 
1460 	now = current_fs_time(inode->i_sb);
1461 	if (!timespec_equal(&inode->i_mtime, &now))
1462 		sync_it = S_MTIME;
1463 
1464 	if (!timespec_equal(&inode->i_ctime, &now))
1465 		sync_it |= S_CTIME;
1466 
1467 	if (IS_I_VERSION(inode))
1468 		sync_it |= S_VERSION;
1469 
1470 	if (!sync_it)
1471 		return;
1472 
1473 	/* Finally allowed to write? Takes lock. */
1474 	if (mnt_want_write_file(file))
1475 		return;
1476 
1477 	/* Only change inode inside the lock region */
1478 	if (sync_it & S_VERSION)
1479 		inode_inc_iversion(inode);
1480 	if (sync_it & S_CTIME)
1481 		inode->i_ctime = now;
1482 	if (sync_it & S_MTIME)
1483 		inode->i_mtime = now;
1484 	mark_inode_dirty_sync(inode);
1485 	mnt_drop_write(file->f_path.mnt);
1486 }
1487 EXPORT_SYMBOL(file_update_time);
1488 
1489 int inode_needs_sync(struct inode *inode)
1490 {
1491 	if (IS_SYNC(inode))
1492 		return 1;
1493 	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1494 		return 1;
1495 	return 0;
1496 }
1497 EXPORT_SYMBOL(inode_needs_sync);
1498 
1499 int inode_wait(void *word)
1500 {
1501 	schedule();
1502 	return 0;
1503 }
1504 EXPORT_SYMBOL(inode_wait);
1505 
1506 /*
1507  * If we try to find an inode in the inode hash while it is being
1508  * deleted, we have to wait until the filesystem completes its
1509  * deletion before reporting that it isn't found.  This function waits
1510  * until the deletion _might_ have completed.  Callers are responsible
1511  * to recheck inode state.
1512  *
1513  * It doesn't matter if I_NEW is not set initially, a call to
1514  * wake_up_inode() after removing from the hash list will DTRT.
1515  *
1516  * This is called with inode_lock held.
1517  */
1518 static void __wait_on_freeing_inode(struct inode *inode)
1519 {
1520 	wait_queue_head_t *wq;
1521 	DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1522 	wq = bit_waitqueue(&inode->i_state, __I_NEW);
1523 	prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1524 	spin_unlock(&inode_lock);
1525 	schedule();
1526 	finish_wait(wq, &wait.wait);
1527 	spin_lock(&inode_lock);
1528 }
1529 
1530 static __initdata unsigned long ihash_entries;
1531 static int __init set_ihash_entries(char *str)
1532 {
1533 	if (!str)
1534 		return 0;
1535 	ihash_entries = simple_strtoul(str, &str, 0);
1536 	return 1;
1537 }
1538 __setup("ihash_entries=", set_ihash_entries);
1539 
1540 /*
1541  * Initialize the waitqueues and inode hash table.
1542  */
1543 void __init inode_init_early(void)
1544 {
1545 	int loop;
1546 
1547 	/* If hashes are distributed across NUMA nodes, defer
1548 	 * hash allocation until vmalloc space is available.
1549 	 */
1550 	if (hashdist)
1551 		return;
1552 
1553 	inode_hashtable =
1554 		alloc_large_system_hash("Inode-cache",
1555 					sizeof(struct hlist_head),
1556 					ihash_entries,
1557 					14,
1558 					HASH_EARLY,
1559 					&i_hash_shift,
1560 					&i_hash_mask,
1561 					0);
1562 
1563 	for (loop = 0; loop < (1 << i_hash_shift); loop++)
1564 		INIT_HLIST_HEAD(&inode_hashtable[loop]);
1565 }
1566 
1567 void __init inode_init(void)
1568 {
1569 	int loop;
1570 
1571 	/* inode slab cache */
1572 	inode_cachep = kmem_cache_create("inode_cache",
1573 					 sizeof(struct inode),
1574 					 0,
1575 					 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1576 					 SLAB_MEM_SPREAD),
1577 					 init_once);
1578 	register_shrinker(&icache_shrinker);
1579 
1580 	/* Hash may have been set up in inode_init_early */
1581 	if (!hashdist)
1582 		return;
1583 
1584 	inode_hashtable =
1585 		alloc_large_system_hash("Inode-cache",
1586 					sizeof(struct hlist_head),
1587 					ihash_entries,
1588 					14,
1589 					0,
1590 					&i_hash_shift,
1591 					&i_hash_mask,
1592 					0);
1593 
1594 	for (loop = 0; loop < (1 << i_hash_shift); loop++)
1595 		INIT_HLIST_HEAD(&inode_hashtable[loop]);
1596 }
1597 
1598 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1599 {
1600 	inode->i_mode = mode;
1601 	if (S_ISCHR(mode)) {
1602 		inode->i_fop = &def_chr_fops;
1603 		inode->i_rdev = rdev;
1604 	} else if (S_ISBLK(mode)) {
1605 		inode->i_fop = &def_blk_fops;
1606 		inode->i_rdev = rdev;
1607 	} else if (S_ISFIFO(mode))
1608 		inode->i_fop = &def_fifo_fops;
1609 	else if (S_ISSOCK(mode))
1610 		inode->i_fop = &bad_sock_fops;
1611 	else
1612 		printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1613 				  " inode %s:%lu\n", mode, inode->i_sb->s_id,
1614 				  inode->i_ino);
1615 }
1616 EXPORT_SYMBOL(init_special_inode);
1617